Early laboratory experiments in biometrics had suggested that the face identification problem was far too complex, even for fantasy computers of the future.  Recognizing faces was a human specialized cognitive ability in the realm of language, creative thought, and human emotions.  When people looked at a face, they didn’t focus so much on shapes or features.  People saw expressions, subtle communication; they experienced past emotions.  People saw the essence of the person in their face.  Two thousand years ago, Cicero said, ‘The face is a picture of the mind as the eyes are its interpreter’.  Today it is said, ‘the eyes are the window to the soul’.  Computers could never duplicate the human experience of recognizing a face.

But it seldom pays to underestimate human ingenuity. 

In 1989, Teuvo Kohonen, a Finnish mathematician and computer scientist specializing in pattern recognition and neural networks, proposed a method for classifying faces derived from the use of a spatial math called eigenvectors that came to be known as ‘eigenfaces’.

Face images were calculated as templates that looked like stick drawings of a face.  The forty or so sticks were ‘elastic’, allowing for infinite template variations.  Differences in templates taken from several pictures of the same face were small when compared to templates calculated from any other face.  This gave a high probability of a match for images of the same face while consistently distinguishing that face from all other faces.

This isn’t anything like the way humans recognize faces, but it works, and by 2006 computers were more accurate at it and much, much faster than people.  In fact, it makes little difference to computer recognition systems if the face has a beard, or is wearing glasses.  Most disguises don’t alter the face geometry that the computer uses to draw its templates, even though they can easily fool people.

Success with object recognition ignited an entire industry.  New, specialized imaging devices dedicated to the strengths of recognition programs vastly improved performance.  Next came face detection, so faces in a crowd could be isolated for recognition processing.